1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-22 18:54:02 +01:00
llvm-mirror/test/CodeGen/AMDGPU/vgpr-liverange.ll
Sebastian Neubauer 5f547ad156 [AMDGPU] Improve killed check for vgpr optimization
The killed flag is not always set. E.g. when a variable is used in a
loop, it is never marked as killed, although it is unused in following
basic blocks. Also, we try to deprecate kill flags and not use them.

Check if the register is live in the endif block. If not, consider it
killed in the then and else blocks.

The vgpr-liverange tests have two new tests with loops
(pre-committed, so the diff is visible).
I also needed to change the subtarget to gfx10.1, otherwise calls
are not working.

Differential Revision: https://reviews.llvm.org/D106291
2021-07-21 15:24:59 +02:00

312 lines
10 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -march=amdgcn -mcpu=gfx1010 -amdgpu-opt-vgpr-liverange=true -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
; a normal if-else
define amdgpu_ps float @else1(i32 %z, float %v) #0 {
; SI-LABEL: else1:
; SI: ; %bb.0: ; %main_body
; SI-NEXT: v_cmp_gt_i32_e32 vcc_lo, 6, v0
; SI-NEXT: ; implicit-def: $vgpr0
; SI-NEXT: s_and_saveexec_b32 s0, vcc_lo
; SI-NEXT: s_xor_b32 s0, exec_lo, s0
; SI-NEXT: ; %bb.1: ; %else
; SI-NEXT: v_mul_f32_e32 v0, 0x40400000, v1
; SI-NEXT: ; implicit-def: $vgpr1
; SI-NEXT: ; %bb.2: ; %Flow
; SI-NEXT: s_or_saveexec_b32 s0, s0
; SI-NEXT: s_xor_b32 exec_lo, exec_lo, s0
; SI-NEXT: ; %bb.3: ; %if
; SI-NEXT: v_add_f32_e32 v0, v1, v1
; SI-NEXT: ; %bb.4: ; %end
; SI-NEXT: s_or_b32 exec_lo, exec_lo, s0
; SI-NEXT: ; return to shader part epilog
main_body:
%cc = icmp sgt i32 %z, 5
br i1 %cc, label %if, label %else
if:
%v.if = fmul float %v, 2.0
br label %end
else:
%v.else = fmul float %v, 3.0
br label %end
end:
%r = phi float [ %v.if, %if ], [ %v.else, %else ]
ret float %r
}
; %v was used after if-else
define amdgpu_ps float @else2(i32 %z, float %v) #0 {
; SI-LABEL: else2:
; SI: ; %bb.0: ; %main_body
; SI-NEXT: v_cmp_gt_i32_e32 vcc_lo, 6, v0
; SI-NEXT: ; implicit-def: $vgpr0
; SI-NEXT: s_and_saveexec_b32 s0, vcc_lo
; SI-NEXT: s_xor_b32 s0, exec_lo, s0
; SI-NEXT: ; %bb.1: ; %else
; SI-NEXT: v_mul_f32_e32 v0, 0x40400000, v1
; SI-NEXT: ; %bb.2: ; %Flow
; SI-NEXT: s_or_saveexec_b32 s0, s0
; SI-NEXT: s_xor_b32 exec_lo, exec_lo, s0
; SI-NEXT: ; %bb.3: ; %if
; SI-NEXT: v_add_f32_e32 v1, v1, v1
; SI-NEXT: v_mov_b32_e32 v0, v1
; SI-NEXT: ; %bb.4: ; %end
; SI-NEXT: s_or_b32 exec_lo, exec_lo, s0
; SI-NEXT: v_add_f32_e32 v0, v1, v0
; SI-NEXT: ; return to shader part epilog
main_body:
%cc = icmp sgt i32 %z, 5
br i1 %cc, label %if, label %else
if:
%v.if = fmul float %v, 2.0
br label %end
else:
%v.else = fmul float %v, 3.0
br label %end
end:
%r0 = phi float [ %v.if, %if ], [ %v, %else ]
%r1 = phi float [ %v.if, %if ], [ %v.else, %else ]
%r2 = fadd float %r0, %r1
ret float %r2
}
; if-else inside loop, %x can be optimized, but %v cannot be.
define amdgpu_ps float @else3(i32 %z, float %v, i32 inreg %bound, i32 %x0) #0 {
; SI-LABEL: else3:
; SI: ; %bb.0: ; %entry
; SI-NEXT: v_cmp_gt_i32_e32 vcc_lo, 6, v0
; SI-NEXT: s_mov_b32 s1, 0
; SI-NEXT: s_branch BB2_2
; SI-NEXT: BB2_1: ; %if.end
; SI-NEXT: ; in Loop: Header=BB2_2 Depth=1
; SI-NEXT: s_or_b32 exec_lo, exec_lo, s2
; SI-NEXT: v_add_nc_u32_e32 v2, 1, v0
; SI-NEXT: s_add_i32 s1, s1, 1
; SI-NEXT: s_cmp_lt_i32 s1, s0
; SI-NEXT: s_cbranch_scc0 BB2_6
; SI-NEXT: BB2_2: ; %for.body
; SI-NEXT: ; =>This Inner Loop Header: Depth=1
; SI-NEXT: ; implicit-def: $vgpr0
; SI-NEXT: ; implicit-def: $vgpr3
; SI-NEXT: s_and_saveexec_b32 s2, vcc_lo
; SI-NEXT: s_xor_b32 s2, exec_lo, s2
; SI-NEXT: ; %bb.3: ; %else
; SI-NEXT: ; in Loop: Header=BB2_2 Depth=1
; SI-NEXT: v_mul_lo_u32 v0, v2, 3
; SI-NEXT: v_mul_f32_e32 v3, v1, v2
; SI-NEXT: ; implicit-def: $vgpr2
; SI-NEXT: ; %bb.4: ; %Flow
; SI-NEXT: ; in Loop: Header=BB2_2 Depth=1
; SI-NEXT: s_or_saveexec_b32 s2, s2
; SI-NEXT: s_xor_b32 exec_lo, exec_lo, s2
; SI-NEXT: s_cbranch_execz BB2_1
; SI-NEXT: ; %bb.5: ; %if
; SI-NEXT: ; in Loop: Header=BB2_2 Depth=1
; SI-NEXT: v_mul_f32_e32 v3, s1, v1
; SI-NEXT: v_add_nc_u32_e32 v0, 1, v2
; SI-NEXT: s_branch BB2_1
; SI-NEXT: BB2_6: ; %for.end
; SI-NEXT: v_add_f32_e32 v0, v0, v3
; SI-NEXT: ; return to shader part epilog
entry:
; %break = icmp sgt i32 %bound, 0
; br i1 %break, label %for.body, label %for.end
br label %for.body
for.body:
%i = phi i32 [ 0, %entry ], [ %inc, %if.end ]
%x = phi i32 [ %x0, %entry ], [ %xinc, %if.end ]
%cc = icmp sgt i32 %z, 5
br i1 %cc, label %if, label %else
if:
%i.tmp = bitcast i32 %i to float
%v.if = fmul float %v, %i.tmp
%x.if = add i32 %x, 1
br label %if.end
else:
%x.tmp = bitcast i32 %x to float
%v.else = fmul float %v, %x.tmp
%x.else = mul i32 %x, 3
br label %if.end
if.end:
%v.endif = phi float [ %v.if, %if ], [ %v.else, %else ]
%x.endif = phi i32 [ %x.if, %if ], [ %x.else, %else ]
%xinc = add i32 %x.endif, 1
%inc = add i32 %i, 1
%cond = icmp slt i32 %inc, %bound
br i1 %cond, label %for.body, label %for.end
for.end:
%x_float = bitcast i32 %x.endif to float
%r = fadd float %x_float, %v.endif
ret float %r
}
; a loop inside an if-else
define amdgpu_ps float @loop(i32 %z, float %v, i32 inreg %bound, float(float)* %extern_func, float(float)* %extern_func2) #0 {
; SI-LABEL: loop:
; SI: ; %bb.0: ; %main_body
; SI-NEXT: v_mov_b32_e32 v6, v0
; SI-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
; SI-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
; SI-NEXT: s_mov_b32 s38, -1
; SI-NEXT: v_mov_b32_e32 v0, v1
; SI-NEXT: v_cmp_gt_i32_e32 vcc_lo, 6, v6
; SI-NEXT: s_mov_b32 s39, 0x31c16000
; SI-NEXT: s_add_u32 s36, s36, s1
; SI-NEXT: s_addc_u32 s37, s37, 0
; SI-NEXT: ; implicit-def: $vgpr1
; SI-NEXT: s_mov_b32 s32, 0
; SI-NEXT: s_and_saveexec_b32 s0, vcc_lo
; SI-NEXT: s_xor_b32 s33, exec_lo, s0
; SI-NEXT: s_cbranch_execz BB3_4
; SI-NEXT: ; %bb.1: ; %else
; SI-NEXT: s_mov_b32 s34, exec_lo
; SI-NEXT: BB3_2: ; =>This Inner Loop Header: Depth=1
; SI-NEXT: v_readfirstlane_b32 s4, v4
; SI-NEXT: v_readfirstlane_b32 s5, v5
; SI-NEXT: v_cmp_eq_u64_e32 vcc_lo, s[4:5], v[4:5]
; SI-NEXT: s_and_saveexec_b32 s35, vcc_lo
; SI-NEXT: s_mov_b64 s[0:1], s[36:37]
; SI-NEXT: s_mov_b64 s[2:3], s[38:39]
; SI-NEXT: s_swappc_b64 s[30:31], s[4:5]
; SI-NEXT: v_mov_b32_e32 v1, v0
; SI-NEXT: ; implicit-def: $vgpr4_vgpr5
; SI-NEXT: ; implicit-def: $vgpr0
; SI-NEXT: s_xor_b32 exec_lo, exec_lo, s35
; SI-NEXT: s_cbranch_execnz BB3_2
; SI-NEXT: ; %bb.3:
; SI-NEXT: s_mov_b32 exec_lo, s34
; SI-NEXT: ; implicit-def: $vgpr0
; SI-NEXT: ; implicit-def: $vgpr2
; SI-NEXT: BB3_4: ; %Flow
; SI-NEXT: s_or_saveexec_b32 s33, s33
; SI-NEXT: s_xor_b32 exec_lo, exec_lo, s33
; SI-NEXT: s_cbranch_execz BB3_8
; SI-NEXT: ; %bb.5: ; %if
; SI-NEXT: s_mov_b32 s34, exec_lo
; SI-NEXT: BB3_6: ; =>This Inner Loop Header: Depth=1
; SI-NEXT: v_readfirstlane_b32 s4, v2
; SI-NEXT: v_readfirstlane_b32 s5, v3
; SI-NEXT: v_cmp_eq_u64_e32 vcc_lo, s[4:5], v[2:3]
; SI-NEXT: s_and_saveexec_b32 s35, vcc_lo
; SI-NEXT: s_mov_b64 s[0:1], s[36:37]
; SI-NEXT: s_mov_b64 s[2:3], s[38:39]
; SI-NEXT: s_swappc_b64 s[30:31], s[4:5]
; SI-NEXT: v_mov_b32_e32 v1, v0
; SI-NEXT: ; implicit-def: $vgpr2_vgpr3
; SI-NEXT: ; implicit-def: $vgpr0
; SI-NEXT: s_xor_b32 exec_lo, exec_lo, s35
; SI-NEXT: s_cbranch_execnz BB3_6
; SI-NEXT: ; %bb.7:
; SI-NEXT: s_mov_b32 exec_lo, s34
; SI-NEXT: BB3_8: ; %end
; SI-NEXT: s_or_b32 exec_lo, exec_lo, s33
; SI-NEXT: v_mov_b32_e32 v0, v1
; SI-NEXT: ; return to shader part epilog
main_body:
%cc = icmp sgt i32 %z, 5
br i1 %cc, label %if, label %else
if:
%v.if = call amdgpu_gfx float %extern_func(float %v)
br label %end
else:
%v.else = call amdgpu_gfx float %extern_func2(float %v)
br label %end
end:
%r = phi float [ %v.if, %if ], [ %v.else, %else ]
ret float %r
}
; a loop inside an if-else, but the variable is still in use after the if-else
define amdgpu_ps float @loop_with_use(i32 %z, float %v, i32 inreg %bound, float(float)* %extern_func, float(float)* %extern_func2) #0 {
; SI-LABEL: loop_with_use:
; SI: ; %bb.0: ; %main_body
; SI-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
; SI-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
; SI-NEXT: s_mov_b32 s38, -1
; SI-NEXT: v_cmp_gt_i32_e32 vcc_lo, 6, v0
; SI-NEXT: v_mov_b32_e32 v40, v1
; SI-NEXT: s_mov_b32 s39, 0x31c16000
; SI-NEXT: s_add_u32 s36, s36, s1
; SI-NEXT: s_addc_u32 s37, s37, 0
; SI-NEXT: ; implicit-def: $vgpr0
; SI-NEXT: s_mov_b32 s32, 0
; SI-NEXT: s_and_saveexec_b32 s0, vcc_lo
; SI-NEXT: s_xor_b32 s33, exec_lo, s0
; SI-NEXT: s_cbranch_execz BB4_4
; SI-NEXT: ; %bb.1: ; %else
; SI-NEXT: s_mov_b32 s34, exec_lo
; SI-NEXT: BB4_2: ; =>This Inner Loop Header: Depth=1
; SI-NEXT: v_readfirstlane_b32 s4, v4
; SI-NEXT: v_readfirstlane_b32 s5, v5
; SI-NEXT: v_cmp_eq_u64_e32 vcc_lo, s[4:5], v[4:5]
; SI-NEXT: s_and_saveexec_b32 s35, vcc_lo
; SI-NEXT: v_mov_b32_e32 v0, v40
; SI-NEXT: s_mov_b64 s[0:1], s[36:37]
; SI-NEXT: s_mov_b64 s[2:3], s[38:39]
; SI-NEXT: s_swappc_b64 s[30:31], s[4:5]
; SI-NEXT: ; implicit-def: $vgpr4_vgpr5
; SI-NEXT: s_xor_b32 exec_lo, exec_lo, s35
; SI-NEXT: s_cbranch_execnz BB4_2
; SI-NEXT: ; %bb.3:
; SI-NEXT: s_mov_b32 exec_lo, s34
; SI-NEXT: ; implicit-def: $vgpr2
; SI-NEXT: BB4_4: ; %Flow
; SI-NEXT: s_or_saveexec_b32 s33, s33
; SI-NEXT: s_xor_b32 exec_lo, exec_lo, s33
; SI-NEXT: s_cbranch_execz BB4_8
; SI-NEXT: ; %bb.5: ; %if
; SI-NEXT: s_mov_b32 s34, exec_lo
; SI-NEXT: BB4_6: ; =>This Inner Loop Header: Depth=1
; SI-NEXT: v_readfirstlane_b32 s4, v2
; SI-NEXT: v_readfirstlane_b32 s5, v3
; SI-NEXT: v_cmp_eq_u64_e32 vcc_lo, s[4:5], v[2:3]
; SI-NEXT: s_and_saveexec_b32 s35, vcc_lo
; SI-NEXT: v_mov_b32_e32 v0, v40
; SI-NEXT: s_mov_b64 s[0:1], s[36:37]
; SI-NEXT: s_mov_b64 s[2:3], s[38:39]
; SI-NEXT: s_swappc_b64 s[30:31], s[4:5]
; SI-NEXT: ; implicit-def: $vgpr2_vgpr3
; SI-NEXT: s_xor_b32 exec_lo, exec_lo, s35
; SI-NEXT: s_cbranch_execnz BB4_6
; SI-NEXT: ; %bb.7:
; SI-NEXT: s_mov_b32 exec_lo, s34
; SI-NEXT: BB4_8: ; %end
; SI-NEXT: s_or_b32 exec_lo, exec_lo, s33
; SI-NEXT: v_add_f32_e32 v0, v0, v40
; SI-NEXT: ; return to shader part epilog
main_body:
%cc = icmp sgt i32 %z, 5
br i1 %cc, label %if, label %else
if:
%v.if = call amdgpu_gfx float %extern_func(float %v)
br label %end
else:
%v.else = call amdgpu_gfx float %extern_func2(float %v)
br label %end
end:
%r = phi float [ %v.if, %if ], [ %v.else, %else ]
%r2 = fadd float %r, %v
ret float %r2
}
attributes #0 = { nounwind }